CUBLAS Flag (`CUBLAS_GEMM_DFALT_TENSOR_OP` -> `CUBLAS_GEMM_DEFAULT_TENSOR_OP`) (#100976)
Looking at the docs https://docs.nvidia.com/cuda/cublas/index.html?highlight=cublasGemmEx#cublasgemmex It seems like the flag should be `CUBLAS_GEMM_DEFAULT_TENSOR_OP`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100976
Approved by: https://github.com/ezyang